Google’s AI Overview feature went viral earlier this week when it generated bizarre responses, such as suggesting users use glue to make cheese sticks on a pizza or advising them to eat rocks.
This new AI feature was designed to improve search results, but when faced with unusual questions, it quickly became a source of internet memes. Google has since attributed these inaccurate responses to "data voids" or "information gaps."
In a blog post on Thursday, Liz Reid, Google’s head of search, acknowledged that AI Overview can "hallucinate" or fabricate information.
She explained that these errors typically stem from "misinterpreting queries, misinterpreting a nuance of language on the web, or not having a lot of great information available."
Reid stated, “We saw AI Overviews that featured sarcastic or troll-y content from discussion forums. Forums are often a great source of authentic, first-hand information, but in some cases can lead to less-than-helpful advice, like using glue to get cheese to stick to pizza.”
Google also clarified that while the inaccurate AI Overview responses were circulating online, many fake screenshots were also being shared.
Notably, AI Overview results for queries about leaving dogs in cars, smoking while pregnant, and depression were never displayed.
Reid admitted that AI Overview struggles with interpreting nonsensical queries and satirical content.
The company has updated its systems "to limit the use of user-generated content in responses that could offer misleading advice" and added "triggering restrictions for queries where AI Overviews were not proving to be as helpful."
Google has acknowledged, “In a small number of cases, we have seen AI Overviews misinterpret language on webpages and present inaccurate information. We worked quickly to address these issues, either through improvements to our algorithms or through established processes to remove responses that don't comply with our policies.”
Going forward, Google AI Overview will not appear for "hard news topics" where "freshness and factuality are important." In terms of health information, Google has implemented "additional triggering refinements to enhance our quality protections."